Skip to content

Latest commit

 

History

History
221 lines (175 loc) · 9.26 KB

index.md

File metadata and controls

221 lines (175 loc) · 9.26 KB
layout title description
default
Customer Segmentation with KMeans Clustering
A short description of your project

Customer Segmentation Project Using K-Means Clustering

Jupyter notebook available here

Table of Contents

  1. Overview
  2. Dataset Description
  3. Objective
  4. Theoretical Background
  5. Methodology
  6. Key Results
  7. Customer Segmentation Strategies
  8. Conclusion
  9. References

Overview

This project utilizes K-Means clustering, an unsupervised machine learning algorithm, to segment customers based on their purchasing behavior. By understanding customer profiles, businesses can tailor their strategies to boost customer satisfaction, loyalty, and revenue.

Dataset Description

The dataset is available here. This Online Retail II data set contains all the transactions occurring for a UK-based and registered, non-store online retail between 01/12/2009 and 09/12/2011.The company mainly sells unique all-occasion gift-ware. Many customers of the company are wholesalers.

Some Additional Information :

Feature Description
InvoiceNo Unique 6-digit identifier for each transaction. Cancellation codes start with 'C'.
StockCode Unique code assigned to each product.
Description Name of the product.
Quantity Number of units purchased in a transaction.
InvoiceDate Date and time of the transaction.
UnitPrice Price per unit in GBP (£).
CustomerID Unique identifier for each customer.
Country Country where the customer resides.

Objective

The primary goals of this project are:

  • To identify meaningful customer groups based on purchasing behavior.
  • To understand the characteristics of each customer segment.
  • To provide actionable insights for improving marketing and customer engagement strategies.

Theoretical Background

Why K-Means Clustering?

K-Means is chosen for this project because:

  1. Efficiency: K-Means works well with large datasets and provides quick clustering results.
  2. Scalability: It can handle a variety of data sizes and complexities.
  3. Interpretability: The results are easy to visualize and interpret, especially for customer segmentation tasks.
  4. Versatility: It works well with numerical data, which is predominant in this dataset.

Understanding the Algorithm

Steps in K-Means Clustering:

  1. Initialization:
    • Choose the number of clusters (‘k’).
    • Randomly initialize cluster centroids.
  2. Assignment Step:
    • Assign each data point to the nearest centroid using the Euclidean distance.
  3. Update Step:
    • Recalculate the centroids by taking the mean of all points assigned to a cluster.
  4. Convergence:
    • Repeat steps 2 and 3 until the centroids no longer change significantly or a predefined number of iterations is reached.

Mathematical Objective:

K-Means minimizes the Within-Cluster Sum of Squares (WCSS):

{% raw %} $$WCSS = \sum_{i=1}^{k} \sum_{x \in C_i} | x - \mu_i |^2$$ {% endraw %}

Where:

  • k: Number of clusters
  • $C_{i}$: Cluster ( i )
  • $\mu_{i}$: Centroid of cluster ( i )
  • x: Data point

Silhouette Score:

The Silhouette Score evaluates the quality of clustering by comparing intra-cluster and inter-cluster distances. It ranges from (-1) to (1):

$$ S(i) = \frac{b(i) - a(i)}{\max{a(i), b(i)}} $$

Where:

  • ( a(i) ): Average distance between ( i ) and other points in the same cluster.
  • ( b(i) ): Average distance between ( i ) and points in the nearest cluster.

Methodology

Exploratory Data Analysis (EDA)

EDA involved:

  1. Checking for null values and invalid data (e.g., negative prices).
  2. Analyzing the distribution of numerical features like Quantity, UnitPrice, and CustomerID.
  3. Identifying patterns in InvoiceDate and Country to uncover insights into purchasing trends.

Observations:

  • A significant number of transactions had missing CustomerID values.
  • Negative or zero values in Quantity and UnitPrice indicated potential errors or special promotions.

Data Cleaning

  1. Removing Invalid Entries:
    • Dropped rows with missing CustomerID.
    • Excluded transactions with negative or zero Quantity and UnitPrice.
  2. Processing Categorical Variables:
    • Cleaned StockCode by removing non-product entries like "ADJUST" and "TEST".

Feature Engineering

  1. Created New Features:
    • sales_line_total = Quantity × UnitPrice: Total revenue per transaction.
  2. Aggregated Data:
    • Grouped by CustomerID to calculate:
      • Monetary Value: Total spending.
      • Frequency: Number of unique transactions.
      • Recency: Days since the last purchase.

Data Scaling and Preprocessing

  • Used StandardScaler to normalize features (Monetary Value, Frequency, Recency) with Z-score scaling:

    $$ Z = \frac{X - \mu}{\sigma} $$

Where :

  • $Z$: The Z-score, representing the normalized value of the feature. It indicates how many standard deviations a particular value $X$ is from the mean $\mu$.

  • $X$: The original value of the feature being normalized.

  • $\mu$: The mean (average) value of the feature, calculated as: $$ \mu = \frac{\sum_{i=1}^{n} X_i}{n} $$ where:

    • $n$: Number of data points.
    • $X_i$: Individual feature values.

  • $\sigma$: The standard deviation of the feature, measuring the spread or variability of the data around the mean, calculated as: $$ \sigma = \sqrt{\frac{\sum_{i=1}^{n} (X_i - \mu)^2}{n}} $$

Clustering Process

  1. Optimal K Selection:

    • Used the Elbow Method and Silhouette Scores to determine $k$ = 4.

  2. K-Means Execution:

    • Clustered scaled data into four segments.
    Elbow Method Graph

Key Results

Cluster Analysis

Cluster Characteristics Description
0 High-value, frequent buyers Regular buyers with high spending.
1 Infrequent, low-value customers Customers who purchase sporadically.
2 New or low-engagement customers Customers with low spending but recent activity.
3 Loyal, high-frequency, high-value buyers The most valuable customers in terms of revenue and engagement.

Customer Segmentation Strategies

Cluster 0: "Retain"

  • Rationale: High-value customers who make frequent purchases.
  • Actions:
    • Introduce loyalty programs.
    • Provide personalized recommendations.

Cluster 1: "Re-Engage"

  • Rationale: Customers with sporadic and low spending patterns.
  • Actions:
    • Send targeted email campaigns.
    • Offer exclusive discounts to incentivize purchases.

Cluster 2: "Nurture"

  • Rationale: Recent customers with potential for growth.
  • Actions:
    • Offer onboarding incentives.
    • Provide exceptional customer service.

Cluster 3: "Reward"

  • Rationale: Top-performing customers in terms of revenue and frequency.
  • Actions:
    • Develop a premium rewards program.
    • Provide exclusive early access to products or events.

Conclusion

K-Means clustering successfully segmented customers into actionable groups. These insights can drive personalized marketing and improve overall business strategy. Elbow Method Graph

Next Steps

  1. Deploy the clustering model in a production environment.
  2. Integrate segmentation results into CRM systems.
  3. Experiment with alternative clustering algorithms (e.g., Hierarchical Clustering).

References